Here is a small collection of .gif files representing recent progress in developing pattern spaces for level-1 trigger applications. This method is based on scaled dimension lowering, as described at the last collaboration meeting. The pattern space parameters are derived from elements of the power spectra of the dimension-lowering distributions.
We are using the same four sets (100 each) of events generated by Ron Longacre - rhic, landau, smoke and chiral. To these we have added a fifth set of 100 Poisson events, each consisting of about 5000 points randomly distributed, to serve as a minimum-correlation reference case.
Figure phcomp -
shows power (inferred anomaly amplitude) vs phase for the five event types and for two approaches to handling the transformation of the dimension-lowering to a power spectrum. One can see that the rhic and Poisson events are at low power, and the three anomalous types are segregated into groups at higher power.
show 3-D representations of a three-phase representation of the events. From the correlations in these figures it is evident that a two-phase vs power representation is possible.
show various representations of two-phase vs power pattern spaces.
Figures snglpk1 and snglpk 2 -
show single detailed plots of the two-phase vs power pattern spaces.
show 2-D phase plots for the two transformation methods. The power degree of freedom is not shown (see snglpk 1 and 2). One sees here very good segregation of types, and in the second method clear separation of event types into subtypes.
______________
These figures are part of work in progress. The pattern spaces shown here have now been incorporated into an event filter, so that we can view the event subensembles that correspond to different parts of the pattern spaces. In this way we map out the correspondence of various pattern-space regions to different event morphologies.
Tom Trainor and Jeff Reid
